Clothing quality inspection task allocation method, system and device based on multi-agent

Abstract

This application discloses a multi-agent-based method, system, and device for assigning garment quality inspection tasks, relating to the field of garment quality inspection technology. The method includes: constructing a quality inspection task agent and a QC agent to characterize the characteristics of the quality inspection task and the capabilities of the quality inspectors, respectively; acquiring multi-dimensional task characteristic data corresponding to each quality inspection task, and generating a comprehensive task score for each quality inspection task based on a preset first evaluation model; acquiring multi-dimensional personal capability data corresponding to each quality inspector, and generating a comprehensive capability score for each quality inspector based on a preset second evaluation model; calculating the task matching degree between each quality inspector and each quality inspection task based on the comprehensive task score, the comprehensive capability score, and a preset dynamic influence factor; dynamically updating the value of the preset dynamic influence factor; and generating and assigning a quality inspection task allocation scheme based on the task matching degree. This application achieves efficient, accurate, and stable allocation of garment quality inspection tasks.

Description

Technical Field

[0001] This application relates to the field of garment quality inspection technology, and in particular to a method, system and device for garment quality inspection task allocation based on multi-agent systems. Background Technology

[0002] As consumer demands become increasingly diversified and personalized, modern apparel companies are continuously expanding their product structures, typically encompassing hundreds of categories and tens of thousands of styles and color combinations. While such highly complex product lines effectively meet the diverse needs of the market, they significantly increase the difficulty of quality management in the production process, especially in the quality inspection stage, which faces multiple challenges such as a surge in task volume, diverse characteristics, and urgent response times.

[0003] Currently, the task allocation for quality control (QC) personnel in the industry still mainly relies on the subjective experience of managers or static scheduling methods based on simple rules, which has the following technical shortcomings: Firstly, in terms of efficiency, manual allocation requires a lot of coordination time, which is difficult to adapt to the flexible production rhythm of high frequency, small batch and fast delivery. Secondly, in terms of accuracy, reasonable task allocation needs to comprehensively consider the real-time working status, skills and expertise, historical performance of each QC, as well as the differentiated characteristics of each quality inspection task in terms of process complexity and urgency. However, manual methods are difficult to fully grasp and dynamically weigh the above multi-dimensional information, which can easily lead to a mismatch between tasks and personnel capabilities, resulting in missed inspections, misjudgments, or rework.

[0004] Furthermore, as businesses expand, quality inspection tasks become increasingly complex. The combinations of design complexity, process difficulty, dyeing uniformity, color difference, and fabric characteristics are countless, and traditional methods lack effective data recording and analysis tools. It is difficult to systematically manage and utilize information such as QC performance, task completion, and historical quality inspection data. This makes it difficult for companies to promptly identify problems in quality inspection work and to conduct scientific evaluation and training of QC personnel, hindering the improvement of the company's overall quality inspection level.

[0005] Therefore, there is an urgent need for an allocation scheme that can integrate multi-source heterogeneous data and has adaptive scheduling capabilities to achieve efficient, accurate and stable allocation of garment quality inspection tasks. Summary of the Invention

[0006] To overcome the shortcomings of existing technologies and provide an allocation scheme that can integrate multi-source heterogeneous data and has adaptive scheduling capabilities to achieve efficient, accurate, and stable allocation of garment quality inspection tasks, this application provides a garment quality inspection task allocation method, system, and device based on multi-agent intelligence.

[0007] Firstly, the objective of this invention is achieved through the following technical solution: A multi-agent-based method for assigning garment quality inspection tasks includes: Construct a quality inspection task agent and a QC agent to represent the characteristics of quality inspection tasks and the capabilities of quality inspection personnel, respectively. Obtain multi-dimensional task characteristic data corresponding to each quality inspection task, and generate a comprehensive task score for each quality inspection task based on the preset first evaluation model; Obtain multi-dimensional personal ability data for each quality inspector, and generate a comprehensive ability score for each quality inspector based on a preset second evaluation model; Based on the overall task score, the overall ability score, and the preset dynamic influence factor, the task matching degree between each quality inspector and each quality inspection task is calculated; the value of the preset dynamic influence factor is dynamically updated according to the timestamp and the degree of work fatigue of the personnel. A quality inspection task allocation plan is generated based on the task matching degree, and the quality inspection task allocation plan is sent to the corresponding quality inspection personnel for execution.

[0008] By adopting the above technical solutions, in order to automate and intelligently allocate quality inspection tasks, this application constructs a quality inspection task agent and a QC agent. This transforms the originally scattered and unstructured task characteristics and personnel capability information into calculable and comparable quantitative representations, eliminating reliance on subjective judgment by managers and significantly improving the automation level of task allocation. To improve the accuracy and adaptability of task allocation, a comprehensive task score and a comprehensive QC capability score are generated based on the first and second evaluation models, respectively. Environmental factors (such as work fatigue levels) that are dynamically updated over time and based on the current working hours of personnel are also introduced. This ensures that the matching degree calculation truly reflects the real-time matching relationship between task difficulty and personnel status, effectively avoiding unreasonable situations such as "assigning high-difficulty tasks to low-ability or high-load personnel." Furthermore, to address the complexity of quality inspection brought about by the diverse product categories and explosive growth in style and color combinations in the apparel industry, this application integrates multi-source heterogeneous data with agent collaboration, possessing good scalability and generalization capabilities to stably handle the high-complexity and high-dynamic quality task scheduling requirements. Because the allocation decision in this application is based on objective data and a pre-set first and second evaluation model, it reduces human intervention and subjective bias, enabling more consistent processing standards for the same or similar tasks across different times and personnel, thereby improving the reliability and traceability of the overall quality inspection results. Therefore, this application not only provides an allocation scheme with adaptive scheduling capabilities but also achieves efficient, accurate, and stable allocation of garment quality inspection tasks.

[0009] In a preferred example of this application, the expression for the first evaluation model is: Where T is the overall task score, and n is the number of characteristic parameters of the quality inspection task. This is the normalized quantified value of the i-th quality inspection task characteristic parameter, which includes style complexity, number of colors and matching difficulty, fabric characteristics, urgency level, and quality risk level. The corresponding first weight, and satisfying .

[0010] By adopting the above technical solution, a weighted summation formula is used to quantify and integrate multi-dimensional task parameters such as style complexity, color matching difficulty, fabric characteristics, urgency, and quality risk level, enabling horizontal comparison of quality inspection tasks of different natures under a unified scale. The first evaluation model not only retains the independent contribution of each task characteristic, but also ensures the stability and interpretability of the evaluation results through normalized weights.

[0011] In a preferred example of this application: the expression for the second evaluation model is: Where Q represents the overall ability score, and m represents the number of individual QC ability parameters. This is the normalized quantified value of the j-th QC individual competency parameter, which includes job rating, quality inspection efficiency, shift scheduling, fatigue index, learning ability, and familiarity with different fabrics. The corresponding second weight, and satisfying .

[0012] By adopting the above technical solution, multiple dimensions of competency indicators, such as job rating, quality inspection efficiency, shift scheduling, fatigue index, learning ability, and fabric familiarity, are integrated into a single comprehensive competency score, enabling dynamic and quantitative assessment of QC personnel competency. The second assessment model breaks through the traditional static evaluation method that relies solely on job level or experience, and can reflect the current state of personnel and skill suitability in real time. In particular, incorporating the "fatigue index" into the competency assessment system allows the system to proactively avoid high-load personnel during allocation, which helps to reduce the risk of human error while ensuring quality inspection quality.

[0013] In a preferred embodiment of this application: the calculation of the task matching degree between each quality inspector and each quality inspection task based on the overall task score, the overall ability score, and a preset dynamic influence factor includes: The task matching degree M is calculated using the following formula: in, The overall score for the current quality inspection task. The average of the overall scores of all pending quality inspection tasks; This is a comprehensive score for the current quality inspectors. S is the average of the comprehensive ability scores of all available quality inspectors; S is the fabric matching degree, which represents the similarity between the fabric type involved in the quality inspection task and the fabric types that the quality inspector has handled in the past, and the value range is [0, 1]; P is the QC work fatigue level, and the value range is [0, 1]. The smaller the value, the lower the fatigue level. The preset adjustment weights, and satisfy the following conditions: .

[0014] By adopting the above technical solutions, a standardized matching formula is introduced to organically integrate the relative suitability of task difficulty and personnel ability, fabric professional matching degree, and real-time fatigue status, forming a multi-factor collaborative decision-making strategy. This effectively avoids deviations caused by matching absolute values, such as forcing high-scoring tasks to be assigned to high-scoring personnel while ignoring the workload. Furthermore, the fabric matching degree S is used to strengthen professional matching, and the physiological state constraint is introduced through the fatigue level P, thereby improving the rationality of the allocation results.

[0015] In a preferred example of this application: the degree of work fatigue of the personnel is characterized by a fatigue index, which is calculated based on the continuous working hours of the quality inspectors, the number of quality inspection tasks completed on the day, and the heart rate variability or eye movement frequency data collected by physiological monitoring sensors, and dynamically recovers over time according to a preset decay function. After generating the quality inspection task allocation scheme, the completion quality score, actual time consumption, and abnormal feedback information during the task execution process are collected to determine the scheme feedback parameters; and the weight parameters in the first evaluation model and the second evaluation model are updated online based on the scheme feedback parameters.

[0016] By adopting the above technical solution, a fatigue index is constructed by integrating continuous working hours, number of completed tasks, and physiological sensor data such as heart rate variability or eye movement frequency. A decay function is introduced to achieve dynamic recovery, transforming "employee work fatigue level" from a subjective estimate into an objective, continuous, and quantifiable technical parameter. Simultaneously, quality scores, time consumption, and anomaly feedback are collected after task execution, and the weight parameters in the evaluation model are iteratively updated online. This forms a closed-loop learning mechanism based on feedback and subsequent optimization, improving the accuracy of fatigue modeling.

[0017] In a preferred example, this application further includes, before generating the quality inspection task allocation scheme: Based on the current production line status, personnel location information, equipment availability, and historical task execution data, a digital twin simulation model for garment quality inspection is constructed. In the digital twin simulation model, multiple candidate allocation schemes are generated for the same set of quality inspection tasks to be assigned. Each candidate scheme corresponds to a different core allocation parameter configuration, and the core allocation parameters include a first weight set. Second weight set Adjusting weights and mandatory allocation thresholds for high-risk tasks; Parallel simulation execution is performed on each candidate allocation scheme to simulate the task completion process and output the corresponding simulation evaluation indicators. The simulation evaluation indicators include the expected total completion time, the expected quality pass rate, the personnel workload balance, and the probability of missing detection of high-risk tasks. Based on the simulation evaluation indicators, a multi-objective optimization algorithm is used to rank the candidate schemes, and the candidate scheme with the best overall performance is selected as the final allocation scheme and executed.

[0018] By adopting the above technical solution, to enhance system robustness and decision-making foresight, this application further employs a digital twin simulation model before task allocation. This model performs parallel simulations on candidate schemes under various core allocation parameter configurations, predicting their performance in dimensions such as estimated completion time, quality pass rate, load balancing, and the probability of missing high-risk tasks. Finally, a multi-objective optimization algorithm is used to select the optimal solution. This upgrades the traditional one-time allocation to the optimal selection verified through simulation, effectively avoiding scheduling failures or resource waste caused by improper parameter configuration. Especially in the face of disruptive scenarios such as emergency order insertions or staff absences, simulations can predict risks and adjust strategies in advance.

[0019] In a preferred embodiment, the implementation of the multi-objective optimization algorithm includes: Construct a task characteristic enhancement matrix that includes task urgency coefficient, quality risk level weight, style complexity sensitivity factor, and fabric scarcity index; The NSGA-II multi-objective genetic algorithm is used to jointly optimize the task characteristic enhancement matrix and core allocation parameters to generate a Pareto front solution set. Optimization preferences are dynamically set based on the company's current production goals: when delivery time is the priority goal, the weight of the estimated total completion time is increased; when zero defects in quality is the priority goal, the weight of the estimated quality pass rate and the probability of missed inspections of high-risk tasks is increased. The optimal solution that meets the current optimization preference is selected from the Pareto front solution set. The core allocation parameter configuration corresponding to the optimal solution is extracted and used to update the weight parameters in the first evaluation model, the second evaluation model and the matching degree calculation model, so as to realize the adaptive evolution of the quality inspection task allocation scheme.

[0020] By adopting the above technical solution, a task characteristic reinforcement matrix is ​​constructed, incorporating factors such as task urgency, quality risk level, style complexity, and fabric scarcity. The NSGA-II algorithm is then used for multi-objective joint optimization to generate a Pareto front solution set. This allows the system to flexibly balance conflicting objectives such as "delivery timeliness" and "zero defects in quality." Enterprises can dynamically adjust their optimization preferences based on their current strategy, automatically selecting the most suitable allocation parameter configuration and updating the evaluation model weights in reverse, achieving adaptive evolution of the allocation strategy. This ensures that quality inspection task allocation truly serves the enterprise's business objectives, significantly improving the flexibility and intelligence of the multi-agent system.

[0021] Secondly, the objective of this invention is achieved through the following technical solution: A multi-agent-based garment quality inspection task allocation system is used to execute the multi-agent-based garment quality inspection task allocation method described above. The system includes: The multi-agent construction module is used to construct quality inspection task agents and QC agents to represent the characteristics of quality inspection tasks and the capabilities of quality inspectors, respectively. The task evaluation module is used to obtain multi-dimensional task characteristic data corresponding to each quality inspection task, and generate a comprehensive task score for each quality inspection task based on the preset first evaluation model. The competency assessment module is used to obtain multi-dimensional personal competency data for each quality inspector and generate a comprehensive competency score for each quality inspector based on a preset second assessment model. The matching degree calculation module is used to calculate the task matching degree between each quality inspector and each quality inspection task based on the task comprehensive score, the comprehensive ability score and the preset dynamic influence factor; the value of the preset dynamic influence factor is dynamically updated according to the timestamp and the degree of work fatigue of the personnel. The task scheduling module is used to generate a quality inspection task allocation plan based on the task matching degree, and send the quality inspection task allocation plan to the corresponding quality inspection personnel for execution; The system is deployed on the server side, stores the status data of the quality inspection task agent and the QC agent in the database, and interfaces with the enterprise production management system to obtain the original task and personnel information.

[0022] Thirdly, the objective of this invention is achieved through the following technical solution: A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described multi-agent-based garment quality inspection task allocation method.

[0023] Fourthly, the objective of this invention is achieved through the following technical solution: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described multi-agent-based garment quality inspection task allocation method.

[0024] In summary, this application includes at least one of the following beneficial technical effects: 1. This application constructs a quality inspection task intelligent agent and a QC intelligent agent, transforming the originally unstructured task characteristics and personnel capabilities into calculable digital representations. Based on the first evaluation model and the second evaluation model, it generates a comprehensive task score and a comprehensive QC capability score, respectively. Combined with environmental factors that are dynamically updated with timestamps and fatigue levels, it calculates the matching degree, and finally realizes the automatic generation of quality inspection task allocation schemes and the automatic allocation of time tasks. 2. It breaks away from the traditional subjective allocation model that relies on human experience, and realizes the transformation from "people finding tasks" to "intelligent matching of tasks and personnel", which significantly improves allocation efficiency and rationality. At the same time, it supports real-time perception of personnel status, enhances the system's adaptability to production fluctuations, and is particularly suitable for high-complexity and high-dynamic quality task scheduling scenarios. Attached Figure Description

[0025] Figure 1 This is a flowchart of a multi-agent-based garment quality inspection task allocation method in one embodiment of this application; Figure 2 This is another flowchart of a multi-agent-based garment quality inspection task allocation method in one embodiment of this application; Figure 3 This is a schematic diagram of a device according to one embodiment of this application. Detailed Implementation

[0026] The present application will be further described in detail below with reference to the accompanying drawings.

[0027] In one embodiment, such as Figure 1 As shown, this application discloses a method for assigning garment quality inspection tasks based on multi-agent systems, which specifically includes the following steps: S1: Construct a quality inspection task agent and a QC agent to represent the characteristics of the quality inspection task and the capabilities of the quality inspection personnel, respectively.

[0028] In this embodiment, the quality inspection task intelligent agent and the QC intelligent agent are set up in a one-to-one correspondence based on the number of quality inspection tasks and the number of QC personnel in actual production. That is, multiple quality inspection task intelligent agents and multiple QC intelligent agents are set up. The quality inspection task intelligent agent refers to a virtual entity that exists in the form of an object or data structure in a computer system. Its attribute fields completely encapsulate all quantifiable features of a garment quality inspection task to be assigned, including the unique task identifier, associated purchase batch number, product category code, style drawing number, color code, fabric composition code, process standard document index, delivery deadline timestamp, and historical defect rate of similar tasks.

[0029] A QC agent is a virtual entity that exists in a computer system as an object or data structure. Its attribute fields fully encapsulate the static qualification information and dynamic status information of a quality inspector, including employee ID, job level, set of skill certification tags, current shift status (on duty / off duty / rest), count of tasks received that day, cumulative value of continuous working hours, quality score of the most recent task completion, and set of fabric types processed in the past.

[0030] For example, before the start of each workday or when a new task arrives, the system retrieves a list of newly added quality inspection tasks from the enterprise's MES system via the task access interface module. Each task record includes the purchase order number PO-20260315-A01, the category as "women's woven shirt", the style number as W-8892, the color as "navy blue (Pantone 19-4023)", the fabric composition as "100% cotton (composition code COT-100)", the process standard referenced as ISO 4915 sewing specification, the delivery date as 18:00 on March 25, 2026, and the historical defect rate for similar products as 2.3%. Based on this, the system instantiates a quality inspection task agent, and the attribute fields are filled according to the preset mapping rules. Simultaneously, the system retrieves the QC personnel list from the HR system and scheduling database. For example, employee ID QC-1078, job level "Senior QC", holds "Woven" and "Knitted" certifications, current status "On Duty", has completed 3 orders that day, worked continuously for 2.5 hours, recent task rating is 4.7 / 5.0, and historically processed fabric types include "COT-100", "POLY-088", and "WOOL-205". Based on this, the system instantiates the QC agent QC-1078. All agents are stored in a PostgreSQL database.

[0031] S2: Obtain multi-dimensional task characteristic data corresponding to each quality inspection task, and generate a comprehensive task score for each quality inspection task based on the preset first evaluation model.

[0032] In this embodiment, multi-dimensional task characteristic data refers to a set of structured values ​​extracted from the attributes of the quality inspection task agent to measure the difficulty and importance of task execution. Specifically, it includes: style complexity, number of colors and matching difficulty, fabric characteristics, urgency, and quality risk level. Style complexity is divided into 1 to 5 levels based on the number of sewing processes in the style drawing. The number of colors and matching difficulty level is as follows: single color = 1, two colors = 2, three colors or more or including gradients = 3. Fabric characteristics are mapped to preset sensitivity levels according to the fabric composition code, such as cotton = 1, silk = 3, functional composite fabric = 4. Urgency is calculated by the remaining hours from the delivery deadline timestamp and the current system time: ≤24h is high = 3, 24~72h is medium = 2, >72h is low = 1. Quality risk level is mapped by the defect rate of similar tasks in history: <1% = 1, 1~3% = 2, >3% = 3.

[0033] The first evaluation model refers to a weighted fusion function with configurable weights deployed on a server, whose input is the normalized value of the above five task characteristic data (linearly mapped to the [0, 1] interval).

[0034] Specifically, the expression for the first evaluation model is: Where T is the overall task score, and n is the number of characteristic parameters of the quality inspection task. This is the normalized quantified value of the i-th quality inspection task characteristic parameter, which includes style complexity, number of colors and matching difficulty, fabric characteristics, urgency level, and quality risk level. The corresponding first weight, and satisfying .

[0035] The characteristic parameters of the quality inspection task refer to a set of discrete or continuous variables used to characterize the internal complexity and external constraints of the clothing quality inspection task, and the number n is fixed at 5. (1) The style complexity is determined according to the number of main sewing processes in the product process sheet. When the number of processes ≤ 5, the value is 1; when 6 - 10, the value is 2; when 11 - 15, the value is 3; when 16 - 20, the value is 4; when > 20, the value is 5. (2) The number of colors and the difficulty of color matching are analyzed according to the color field in the design BOM. For single color, the value is 1; for two colors, the value is 2; for three or more colors or designs including gradients / collisions, the value is 3. (3) The fabric characteristics are found in the preset mapping table according to the fabric composition code. For pure cotton (COT - 100) and polyester (POLY - 088), the value is 1; for blended fabrics, the value is 2; for silk (SILK - 301) and wool (WOOL - 205), the value is 3; for functional composite fabrics, the value is 4. (4) The urgency is calculated by the remaining hours H between the task delivery deadline timestamp and the current system time. If H ≤ 24, the value is 3; if 24 < H ≤ 72, the value is 2; if H > 72, the value is 1. (5) The quality risk level is determined according to the historical defect rate R of similar tasks in the same category within the past 90 days of the procurement batch. If R < 1.0%, the value is 1; if 1.0% ≤ R < 3.0%, the value is 2; if R ≥ 3.0%, the value is 3.

[0036] Specifically, after receiving a new quality inspection task, the system automatically retrieves the associated process sheet, BOM table, delivery schedule, and historical quality database records from the enterprise PLM system. For example, the process sheet corresponding to task PO - 20260315 - A01 shows that there are 14 main sewing processes in total, so the style complexity = 3; the color field in the BOM table is "navy (Pantone 19 - 4023)" without multi - color description, so the number of colors and the difficulty of color matching = 1; the fabric composition is "100% cotton", with the code COT - 100, and looking up the table, the fabric characteristics = 1; the delivery deadline is 18:00 on March 25, 2026, and the current time is 14:00 on March 23, 2026, with 52 hours remaining, so the urgency = 2; querying the historical database, the defect rate of this category (women's woven shirts) in the past 90 days is 2.3%, so the quality risk level = 2. Thus, the original parameter vector is [3, 1, 1, 2, 2].

[0037] Normalized quantization value refers to the standardized numerical value obtained by linearly mapping the above - mentioned original parameter values to the interval [0, 1]. The mapping rule is: for parameters with a value range of , their normalized value . Among them, the style complexity ; the number of colors and the difficulty of color matching ; the fabric characteristics ; the urgency ; the quality risk level .

[0038] First weight set It is a set of preset non-negative real numbers, for example, the specific values ​​are: (Weight of style complexity) (Color difficulty weight) (Weight of fabric characteristics) (Urgency weighting) (Quality risk level weights). This weight set was determined by the Quality Management Expert Committee using the Analytic Hierarchy Process (AHP).

[0039] S3: Obtain multi-dimensional personal ability data for each quality inspector and generate a comprehensive ability score for each quality inspector based on the preset second evaluation model.

[0040] In this embodiment, the QC individual competency parameters are a set of structured variables used to characterize the current competency status of quality inspectors. The number of these variables, m, is fixed at 6, and specifically includes: (1) Job rating: determined according to the job level field in the human resources system, with a value of 1 for junior QC, 2 for intermediate QC, and 3 for senior QC; (2) Quality inspection efficiency: Calculate the average processing time of all quality inspection tasks completed by the QC in the past 7 calendar days. The unit is minutes, and then the formula is used: The original efficiency ratio is calculated. If the ratio is greater than 2.0, it is truncated to 2.0. Finally, it is mapped to the [0, 1] interval and linear compression is applied. ; (3) Scheduling status: The current status is read in real time from the scheduling system. If the status is "on duty", the value is 1; if the status is "off duty", "on leave", or "training", the value is 0. (4) Fatigue index: determined by continuous working hours (Unit: hours) and the number of tasks completed that day Calculate together, the formula is: The results are normalized and restricted to the interval [0, 1]. (5) Learning ability: Based on the quality score sequence of the new fabric type task encountered by QC for the first time in the past 30 days, the slope of the linear regression is fitted k. If k≥0.02, the value is 1; if 0≤k<0.02, the value is 0.5; if k<0, the value is 0. (6) Familiarity with different fabrics: For the fabric composition code involved in the current quality inspection task to be evaluated, check whether the same fabric code is included in the QC historical task record. If it exists, take the value 1; otherwise, take the value 0.

[0041] Specifically, the expression for the second evaluation model is: Where Q represents the overall ability score, and m represents the number of individual QC ability parameters. This is the normalized quantified value of the j-th QC individual competency parameter, which includes job rating, quality inspection efficiency, shift scheduling, fatigue index, learning ability, and familiarity with different fabrics. The corresponding second weight, and satisfying .

[0042] Specifically, job ratings (ranging from 1 to 3) are normalized to... Quality inspection efficiency, shift scheduling, fatigue index, learning ability, and familiarity level are already in the range [0, 1] or can be easily mapped into this interval, so they are directly used as... to Use without additional scaling. Second weight set. It is a set of preset non-negative real numbers, for example, the specific values ​​are: (Job rating weight) Quality inspection efficiency weighting) (Weight of scheduling) (Fatigue index weighting, but included in the negative impact) (Learning ability weight) (Fabric familiarity weight). The second weight set is determined by the quality management department based on the correlation analysis between historical task success rate and personnel performance, and is fixed in the system configuration file.

[0043] S4: Based on the overall task score, overall ability score, and preset dynamic influence factors, calculate the task matching degree between each quality inspector and each quality inspection task; the value of the preset dynamic influence factors is dynamically updated according to the timestamp and the degree of work fatigue of the personnel.

[0044] In this embodiment, the preset dynamic influence factor refers to a correction coefficient that changes continuously with the system running time. The value is determined by the current system timestamp (accurate to the second) and the fatigue index in the QC agent. Specifically, every 10 minutes, if the QC does not rest, the fatigue index automatically increases by 0.02 according to the exponential decay function; if a rest check-in event is detected, it is reset to 0.1.

[0045] Specifically, the task matching degree M is calculated using the following formula: in, The overall score for the current quality inspection task; This is the average of the overall scores for all pending quality inspection tasks. This refers to all unassigned tasks in the system during the current scheduling period. The arithmetic mean of the values; The difficulty level of the task relative to the current task pool (>1 indicates high difficulty, <1 indicates low difficulty). This represents the overall competence score of current quality inspectors. This is the average of the comprehensive competence scores of all available quality control personnel, referring to the scores of all QC personnel whose status is "on duty" and who have not been assigned tasks during the current scheduling cycle. The arithmetic mean of the values; S is the fabric matching degree, which represents the similarity between the fabric type involved in the quality inspection task and the fabric types previously handled by the quality inspector, with a value range of [0, 1]; if the fabric composition code involved in the current quality inspection task exists in the set of fabric types previously handled by the QC personnel, then S = 1.0, otherwise S = 0.0. P is the QC work fatigue level, with a value range of [0, 1], the smaller the value, the lower the fatigue level; The preset adjustment weights, and satisfy the following conditions: For example, .

[0046] For example, the system pairs each task to be assigned with each available QC and calculates the task matching degree. The task matching degree calculation logic is as follows: First, obtain the task's comprehensive score T and the QC's comprehensive ability score Q; then read the current system timestamp (e.g., 2026-03-23 ​​14:30:22) and the QC's fatigue index (e.g., 0.35); the dynamic impact factor is the complement of the fatigue index (1 - fatigue index = 0.65).

[0047] Furthermore, in another embodiment, the degree of employee work fatigue is characterized by a fatigue index, which is calculated based on the continuous working hours of the quality inspectors, the number of quality inspection tasks completed that day, and the heart rate variability or eye movement frequency data collected by physiological monitoring sensors, and dynamically recovers over time according to a preset decay function.

[0048] Continuous working hours The first time stamp of this shift is obtained from the scheduling system and access control clock-in records. The difference between this time and the current system time is calculated and the result is expressed in hours. Number of quality inspection tasks completed on that day ; Count the total number of tasks with a "completed" status for that QC on that day from the MES task execution log; (3) Physiological monitoring sensor data: collected by wearable devices or non-contact cameras deployed at the quality inspection station, including heart rate variability (HRV) and eye movement frequency. Heart rate variability (HRV) is sampled every 5 minutes using a PPG photoelectric sensor, and the standard deviation of adjacent RR intervals (SDNN) is calculated. If SDNN < 30ms, it is judged as a state of high fatigue. Eye movement frequency is tracked by an infrared camera to track the blinking frequency. If the number of blinks per unit time is < 8 times / minute, it is judged as a decrease in attention.

[0049] The system converts the above signals into fatigue contribution values: continuous working time contribution. Task quantity contribution Physiological signal contribution =0.7 (if either HRV or eye movement is abnormal) or 0.3 (if both are normal). Final initial fatigue index value. .

[0050] The preset decay function refers to the automatic recovery of the fatigue index according to the exponential decay law when the system detects that the QC has entered a rest state (such as clocking out or having no operation for 10 consecutive minutes). The formula is:

[0051] in, The fatigue index is the initial value at the start of the rest period, t is the duration of rest (in minutes), and the decay coefficient λ = 0.08, ensuring that the fatigue level drops to about 10% of the initial value after 30 minutes of rest.

[0052] S5: Generate a quality inspection task allocation plan based on the task matching degree, and send the quality inspection task allocation plan to the corresponding quality inspection personnel for execution.

[0053] In this embodiment, the quality inspection task allocation scheme refers to a one-to-one allocation mapping table from a task to a QC, ensuring that each task is assigned to only one QC, and that each QC does not exceed its maximum concurrent task count (default is 1) at the same time. Quality inspection task instructions can be sent via WeChat or MES terminal push, including the task ID, product image, quality inspection standard link, and deadline, and the task is marked as "assigned" in the QC agent.

[0054] Specifically, the system uses a greedy algorithm to generate an allocation scheme: all tasks to be allocated are sorted in descending order of task urgency; for each task, the QC with the highest matching degree and currently not assigned a task is selected from its candidate QC list; if no QC is available, it enters the waiting queue.

[0055] For example, the task PO-20260315-A01 has the highest matching degree with QC-1078 (M=0.502), and QC-1078 currently has no tasks. The system assigns it to QC-1078. The assignment result is written to the assignment_plan table, and a message is sent to QC-1078 via the Enterprise WeChat robot through the API: "You have a new quality inspection task: PO-20260315-A01 (Women's Shirt - Navy Blue), please complete it before 18:00 on March 25th. See the MES terminal for details." At the same time, the QC agent status is updated to "busy", and the task agent status is updated to "assigned".

[0056] Furthermore, after generating the quality inspection task allocation plan, the completion quality score, actual time consumption, and abnormal feedback information during the task execution process are collected to determine the plan feedback parameters; and the weight parameters in the first evaluation model and the second evaluation model are updated online based on the plan feedback parameters.

[0057] Specifically, complete the quality scoring. A score of 0-5 given by the final inspector or AI visual inspection system. Actual time consumed. The time difference between task assignment and QC clicking the "Complete" button. Anomaly feedback information includes Boolean flags such as "Report Missed Inspection," "Rework Mark," and "Associated with Customer Complaint." If any of these are true, the anomaly flag e=1; otherwise, e=0.

[0058] For example, the system constructs a loss function based on feedback parameters and uses mini-batch gradient descent to adjust the weights of the first evaluation model. Weights of the second evaluation model Perform an online update. The specific update rules are as follows: like If e=1, it is judged as "assignment failure", and the weight of underestimated parameters in the task (such as quality risk level) is increased. ) and the overestimation of personnel competencies (such as fabric familiarity) ); like Then increase the weights related to task complexity ( , After accumulating 100 valid feedback samples, a weight fine-tuning is triggered, with an adjustment step size η=0.01. The new weight = old weight + η×Δ, where Δ is the correction vector calculated based on the feedback direction.

[0059] In one embodiment, such as Figure 2 As shown, before generating the quality inspection task allocation scheme, the multi-agent-based garment quality inspection task allocation method also includes: S501: Based on the current production line status, personnel location information, equipment availability, and historical task execution data, construct a digital twin simulation model for garment quality inspection.

[0060] In this embodiment, the digital twin simulation model for garment quality inspection is a discrete event simulation system deployed on a server cluster. Its state is synchronized with the physical production line in real time. The core data sources include the current production line status, personnel location information, equipment availability, and historical task execution data. The current production line status is obtained from the MES system, showing the occupancy status of each quality inspection station and the operating status of equipment (such as whether lighting, magnifying glasses, and color charts are available). Personnel location information is obtained through UWB positioning tags or Wi-Fi fingerprint positioning, acquiring the current workstation coordinates (x, y) of each QC personnel with an accuracy of ±0.5 meters. Equipment availability is read from the IoT device management platform, showing the online status and reservation queue of dedicated testing equipment (such as colorimeters and strength testers). Historical task execution data is loaded from the database, containing records of all quality inspection tasks from the past 180 days, including task type, assigned personnel, actual time consumed, quality score, and anomaly type.

[0061] Specifically, the digital twin simulation model for garment quality inspection is built using the AnyLogic platform. It uses agents as the basic unit, with each quality inspection task agent and QC agent having a corresponding virtual copy in the simulation environment. Their behavior rules are driven by a state machine. The simulation time step is set to 1 second, supporting parallel execution of multiple scenarios.

[0062] S502: In the digital twin simulation model, multiple candidate allocation schemes are generated for the same set of quality inspection tasks to be assigned. Each candidate scheme corresponds to a different core allocation parameter configuration, including the first weight set. Second weight set Adjusting weights And a threshold for mandatory allocation of high-risk tasks.

[0063] In this embodiment, the first weight set These are the 5-dimensional weights used in the first evaluation model to calculate the overall task score, and they are allowed to fluctuate within a preset range, such as... ∈[0.25, 0.35], second weight set These are the 6-dimensional weights used in the second assessment model to calculate the QC capability score, and they can also be adjusted within a reasonable range. The mandatory allocation threshold for high-risk tasks refers to the fact that when the task quality risk level is ≥3, it can only be assigned to QCs with a work rating of 3 and 7≥Q′. This threshold can be set to be turned on / off or the lower limit of Q′ can be adjusted.

[0064] Specifically, Latin hypercube sampling (LHS) is used to generate 20 different configurations in the aforementioned parameter space, each corresponding to a candidate allocation scheme. For each scheme, the simulation engine performs a complete task allocation and execution simulation, outputting four simulation evaluation metrics: (1) The estimated total completion time refers to the simulation time from the current moment to the completion of the last task; (2) The expected quality pass rate is calculated as follows: High-risk tasks refer to those with a quality risk level of 3, while medium-risk tasks refer to those with a quality risk level of 2. (3) The personnel load balance is used to calculate the standard deviation of the expected working hours of each QC. The smaller the value, the more balanced the load. (4) The probability of missing detection for high-risk tasks is the percentage of times that tasks with a quality risk level ≥3 are missed. If there are no high-risk tasks, it is set to 0. If it is assigned to QC with Q′<0.6, the probability of missing detection is set to 0.3. Otherwise, it is 0.05. The total risk value is obtained by summing them up.

[0065] Subsequently, the system uses a weighted summation multi-objective scoring method to rank the 20 solutions: the current production goal is set as "balancing delivery and quality", so the weights are [0.3 (time), 0.4 (pass rate), 0.2 (balance), 0.1 (risk of missed inspection)]. After normalizing each indicator, the weighted summation is performed, and the solution with the highest score is the optimal solution.

[0066] S503: Perform parallel simulation execution of each candidate allocation scheme, simulate the task completion process, and output the corresponding simulation evaluation indicators, including the expected total completion time, expected quality pass rate, personnel workload balance, and probability of missed detection in high-risk tasks.

[0067] In this embodiment, the candidate allocation scheme refers to multiple possible combinations of task-QC mappings generated by the system before the formal allocation, and each scheme corresponds to a complete allocation decision.

[0068] Specifically, parallel simulation execution refers to simultaneously submitting all candidate solutions to simulation engine instances deployed on a Kubernetes cluster. Each instance independently runs a discrete event simulation process, simulating the entire task execution process over the next four hours from the current moment. The simulation process strictly follows these rules: Each QC can only process one task at a time. The task processing time is jointly determined by the task's overall score T and the overall ability score Q assigned to the QC, as shown in the formula: This represents the historical average time taken for tasks in this product category. If the task quality risk level is ≥3 and it is assigned to a QC with Q<0.6, then a "missed detection event" will be triggered with a 30% probability in the simulation. QC movement time is calculated based on the distance between its current position and the task station, with the speed set to 1 meter / second.

[0069] S504: Based on simulation evaluation metrics, a multi-objective optimization algorithm is used to rank the candidate schemes, and the candidate scheme with the best overall performance is selected as the final allocation scheme and executed.

[0070] In this embodiment, the multi-objective optimization algorithm employs a weighted normalized scoring method. Its process is as follows: First, the four simulation evaluation indicators are normalized to uniformly map them to the interval [0, 1], where: The estimated total completion time, personnel workload balance, and probability of missed detections in high-risk tasks are cost-related indicators (the lower the better), and the normalization formula is: ; The expected quality pass rate is a benefit-oriented indicator (the higher the better), and the normalization formula is: ,in Prevent division by zero.

[0071] Subsequently, an optimization preference weight vector is set based on the company's current production strategy. ,satisfy The default configuration is: =0.3 (Timeliness Priority) =0.4 (quality priority) =0.2 (Balanced Priority) =0.1 (risk control priority). When an "urgent delivery" instruction is received, the system automatically switches to w=[0.6, 0.2, 0.1, 0.1]; when entering "Quality Month", it switches to... =[0.2, 0.5, 0.2, 0.1].

[0072] Ultimately, the overall score for each candidate solution is: in, These are the normalized values ​​of the four indicators. The system selects the scheme with the highest score as the final allocation scheme and calls the task scheduling module to send it to the corresponding QC terminal device.

[0073] Specifically, the implementation of multi-objective optimization algorithms includes: S100: Construct a task characteristic enhancement matrix that includes task urgency coefficient, quality risk level weight, style complexity sensitivity factor, and fabric scarcity index.

[0074] In this embodiment, the task characteristic enhancement matrix is ​​a K×4 real number matrix (K is the number of tasks to be assigned at present). Each row corresponds to a task, and the four columns represent the task urgency coefficient, quality risk level weight, style complexity sensitivity factor, and fabric scarcity index, respectively.

[0075] The urgency factor of the task is calculated from the remaining delivery hours H, using the following formula: When H≤24 When H>96 The quality risk level weight is obtained by directly taking the task's quality risk level value (1 / 2 / 3) and normalizing it. The style complexity sensitivity factor is taken from the original style complexity value (1~5) and linearly mapped to [0,1]. The fabric scarcity index is determined by querying a preset scarcity table based on the fabric composition code: pure cotton = 0.1, polyester = 0.1, blended fabrics = 0.3, silk or wool = 0.6, and functional composite fabrics = 0.9.

[0076] For example, the system has 5 tasks to be assigned. Task T1 has a remaining delivery time of 20 hours, a quality risk level of 2, a style complexity of 4, and a fabric value of COT-100. Its reinforcement vector is: Urgency coefficient = 1.0 (since H = 20 ≤ 24), quality risk level weight = (2 − 1) / 2 = 0.5, style complexity sensitivity factor = (4 − 1) / 4 = 0.75, fabric scarcity index = 0.1. Similarly, the remaining tasks are calculated to obtain the task characteristic enhancement matrix. .

[0077] S200: The NSGA-II multi-objective genetic algorithm is used to jointly optimize the task characteristic enhancement matrix and core allocation parameters to generate the Pareto front solution set.

[0078] In this embodiment, the core allocation parameters refer to a set of adjustable strategy parameters, specifically including: The 5-dimensional weights of the first evaluation model ,satisfy and The 6-dimensional weights of the second evaluation model ,satisfy and ; 3D adjustment weights for matching degree calculation ,satisfy and High-risk tasks are forcibly assigned a threshold θ∈[0.60, 0.85].

[0079] Specifically, the NSGA-II multi-objective genetic algorithm is configured to optimize the following three objective functions: Objective 1 (Minimize): Estimated total completion time ; Objective 2 (Maximize): Expected quality pass rate ; Objective 3 (Minimize): Probability of missed detections in high-risk tasks .

[0080] After the NSGA-II multi-objective genetic algorithm finishes running, it outputs a set of non-dominated solutions, also known as the Pareto front solution set, which typically contains 8 to 15 solutions. Each solution contains a complete parameter configuration and the corresponding three-objective values.

[0081] The NSGA-II multi-objective genetic algorithm's parameter settings include: population size = 50, maximum number of generations = 100, crossover probability = 0.9, and mutation probability = 0.1. The crossover operator uses simulated binary crossover (SBX), and the mutation operator uses polynomial mutation, with a crossover distribution exponent of 20 and a mutation distribution exponent of 20. Each individual is encoded as a 20-dimensional real vector (5+6+3+1=15 dimensions, the remaining 5 dimensions being redundant check dimensions), and constraint repair ensures the weight sum is 1. For example, the encoding is a real vector, and each individual contains the complete core assignment parameter configuration:

[0082] Where θ is the mandatory assignment threshold for high-risk tasks, with a value range of [0.6, 0.85].

[0083] S300: Dynamically set optimization preferences based on the company's current production goals: When delivery time is the priority goal, increase the weight of the estimated total completion time; when zero defects in quality is the priority goal, increase the weight of the estimated quality pass rate and the probability of missed inspections of high-risk tasks.

[0084] In this embodiment, the system determines the current optimization preference by reading the global policy flag of the enterprise production management system: if the flag is "EXPEDITE_DELIVERY", it is determined that the delivery time is the priority target, and the estimated total completion time is assigned in the solution selection stage. Higher decision weight; if the flag is "ZERO_DEFECT_QUALITY", it is determined that zero defects in quality is the priority goal, and the expected quality pass rate is increased. High-risk task false negative probability The weights are determined by the system; by default, a balanced weighting is used.

[0085] Specifically, the weight configuration rules are as follows: When delivery timeliness is prioritized, a preference weight vector is set. ; When zero defects are prioritized, λ = [0.2, 0.6, 0.2]; In the equilibrium mode, λ = [0.3, 0.5, 0.2].

[0086] S400: Select the optimal solution that meets the current optimization preference from the Pareto front solution set, extract the core allocation parameter configuration corresponding to the optimal solution, and use it to update the weight parameters in the first evaluation model, the second evaluation model and the matching degree calculation model to realize the adaptive evolution of the quality inspection task allocation scheme.

[0087] In this embodiment, the optimal solution is selected using a weighted normalization scoring method: first, the three metric values ​​of all solutions in the Pareto front are normalized (cost-related metrics are normalized using inverse normalization, and benefit-related metrics are normalized using forward normalization), and then the comprehensive score is calculated:

[0088] Choose the solution with the highest score as the optimal solution.

[0089] Subsequently, the system performs an atomization parameter update operation: Write ω from the optimal solution into the database table task_weight_config; Write v to qc_capability_weight_config; Write β into matching_beta_config; Write θ into high_risk_threshold.

[0090] All write operations are completed within a single database transaction, ensuring consistency. After the update, all newly arriving task assignment requests will immediately use the new parameters.

[0091] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0092] In one embodiment, a multi-agent-based garment quality inspection task allocation system is provided, which corresponds to the multi-agent-based garment quality inspection task allocation method described in the above embodiment.

[0093] The garment quality inspection task allocation system based on multi-agent technology includes a multi-agent construction module, a task evaluation module, a capability evaluation module, a matching degree calculation module, and a task scheduling module. Detailed descriptions of each functional module are as follows: The multi-agent construction module is used to construct quality inspection task agents and QC agents to represent the characteristics of quality inspection tasks and the capabilities of quality inspectors, respectively. The task evaluation module is used to obtain multi-dimensional task characteristic data corresponding to each quality inspection task, and generate a comprehensive task score for each quality inspection task based on the preset first evaluation model. The competency assessment module is used to obtain multi-dimensional personal competency data for each quality inspector and generate a comprehensive competency score for each quality inspector based on a preset second assessment model. The matching degree calculation module is used to calculate the task matching degree between each quality inspector and each quality inspection task based on the task comprehensive score, comprehensive ability score and preset dynamic influence factor; the value of the preset dynamic influence factor is dynamically updated according to the timestamp and the degree of work fatigue of the personnel. The task scheduling module is used to generate a quality inspection task allocation plan based on the task matching degree, and send the quality inspection task allocation plan to the corresponding quality inspection personnel for execution; The system is deployed on the server side, stores the status data of quality inspection task agents and QC agents through a database, and interfaces with the enterprise production management system to obtain original task and personnel information.

[0094] For specific limitations regarding the multi-agent-based garment quality inspection task allocation system, please refer to the limitations of the multi-agent-based garment quality inspection task allocation method mentioned above, which will not be repeated here. Each module in the multi-agent-based garment quality inspection task allocation system can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in the processor of the computer device in hardware form or independent of it, or it can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0095] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores quality inspection task agents, QC agents, etc. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a multi-agent-based garment quality inspection task allocation method.

[0096] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a multi-agent-based garment quality inspection task allocation method.

[0097] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of a multi-agent-based garment quality inspection task allocation method.

[0098] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0099] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0100] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for assigning garment quality inspection tasks based on multi-agent systems, characterized in that, include: Construct a quality inspection task agent and a QC agent to represent the characteristics of quality inspection tasks and the capabilities of quality inspection personnel, respectively. Obtain multi-dimensional task characteristic data corresponding to each quality inspection task, and generate a comprehensive task score for each quality inspection task based on the preset first evaluation model; Obtain multi-dimensional personal ability data for each quality inspector, and generate a comprehensive ability score for each quality inspector based on a preset second evaluation model; Based on the overall task score, the overall ability score, and the preset dynamic influence factor, the task matching degree between each quality inspector and each quality inspection task is calculated; the value of the preset dynamic influence factor is dynamically updated according to the timestamp and the degree of work fatigue of the personnel. A quality inspection task allocation plan is generated based on the task matching degree, and the quality inspection task allocation plan is sent to the corresponding quality inspection personnel for execution.

2. The garment quality inspection task allocation method based on multi-agent technology according to claim 1, characterized in that, The expression for the first evaluation model is: Where T is the overall task score, and n is the number of characteristic parameters of the quality inspection task. This is the normalized quantified value of the i-th quality inspection task characteristic parameter, which includes style complexity, number of colors and matching difficulty, fabric characteristics, urgency level, and quality risk level. The corresponding first weight, and satisfying .

3. The garment quality inspection task allocation method based on multi-agent technology according to claim 1 or 2, characterized in that, The expression for the second evaluation model is: Where Q represents the overall ability score, and m represents the number of individual QC ability parameters. This is the normalized quantified value of the j-th QC individual competency parameter, which includes job rating, quality inspection efficiency, shift scheduling, fatigue index, learning ability, and familiarity with different fabrics. The corresponding second weight, and satisfying .

4. The garment quality inspection task allocation method based on multi-agent technology according to claim 1, characterized in that, The process of calculating the task matching degree between each quality inspector and each quality inspection task based on the overall task score, the overall ability score, and a preset dynamic influence factor includes: The task matching degree M is calculated using the following formula: in, The overall score for the current quality inspection task. The average of the overall scores of all pending quality inspection tasks; This is a comprehensive score for the current quality inspectors. is the average of the comprehensive ability scores of all available quality inspectors; S is the fabric matching degree, which represents the similarity between the fabric type involved in the quality inspection task and the fabric types that the quality inspector has handled in the past, and the value range is [0, 1]; P is the QC work fatigue level, and the value range is [0, 1]. The smaller the value, the lower the fatigue level. The preset adjustment weights, and satisfy the following conditions: .

5. The garment quality inspection task allocation method based on multi-agent technology according to claim 1, characterized in that, The degree of work fatigue of the personnel is characterized by a fatigue index, which is calculated based on the continuous working hours of the quality inspectors, the number of quality inspection tasks completed on the day, and the heart rate variability or eye movement frequency data collected by physiological monitoring sensors, and dynamically recovers over time according to a preset decay function. After generating the quality inspection task allocation scheme, the completion quality score, actual time consumption, and abnormal feedback information during the task execution process are collected to determine the scheme feedback parameters; and the weight parameters in the first evaluation model and the second evaluation model are updated online based on the scheme feedback parameters.

6. The garment quality inspection task allocation method based on multi-agent technology according to claim 1, characterized in that, Before generating the quality inspection task allocation plan, the following steps are also included: Based on the current production line status, personnel location information, equipment availability, and historical task execution data, a digital twin simulation model for garment quality inspection is constructed. In the digital twin simulation model, multiple candidate allocation schemes are generated for the same set of quality inspection tasks to be assigned. Each candidate scheme corresponds to a different core allocation parameter configuration, and the core allocation parameters include a first weight set. Second weight set Adjusting weights and mandatory allocation thresholds for high-risk tasks; Parallel simulation execution is performed on each candidate allocation scheme to simulate the task completion process and output the corresponding simulation evaluation indicators. The simulation evaluation indicators include the expected total completion time, the expected quality pass rate, the personnel workload balance, and the probability of missing detection of high-risk tasks. Based on the simulation evaluation indicators, a multi-objective optimization algorithm is used to rank the candidate schemes, and the candidate scheme with the best overall performance is selected as the final allocation scheme and executed.

7. The garment quality inspection task allocation method based on multi-agent technology according to claim 6, characterized in that, The implementation of the multi-objective optimization algorithm includes: Construct a task characteristic enhancement matrix that includes task urgency coefficient, quality risk level weight, style complexity sensitivity factor, and fabric scarcity index; The NSGA-II multi-objective genetic algorithm is used to jointly optimize the task characteristic enhancement matrix and core allocation parameters to generate a Pareto front solution set. Optimization preferences are dynamically set based on the company's current production goals: when delivery time is the priority goal, the weight of the estimated total completion time is increased; when zero defects in quality is the priority goal, the weight of the estimated quality pass rate and the probability of missed inspections of high-risk tasks is increased. The optimal solution that meets the current optimization preference is selected from the Pareto front solution set. The core allocation parameter configuration corresponding to the optimal solution is extracted and used to update the weight parameters in the first evaluation model, the second evaluation model and the matching degree calculation model, so as to realize the adaptive evolution of the quality inspection task allocation scheme.

8. A multi-agent-based garment quality inspection task allocation system, characterized in that, The system is used to execute the multi-agent-based garment quality inspection task allocation method as described in any one of claims 1-7, the system comprising: The multi-agent construction module is used to construct quality inspection task agents and QC agents to represent the characteristics of quality inspection tasks and the capabilities of quality inspectors, respectively. The task evaluation module is used to obtain multi-dimensional task characteristic data corresponding to each quality inspection task, and generate a comprehensive task score for each quality inspection task based on the preset first evaluation model. The competency assessment module is used to obtain multi-dimensional personal competency data for each quality inspector and generate a comprehensive competency score for each quality inspector based on a preset second assessment model. The matching degree calculation module is used to calculate the task matching degree between each quality inspector and each quality inspection task based on the task comprehensive score, the comprehensive ability score and the preset dynamic influence factor; the value of the preset dynamic influence factor is dynamically updated according to the timestamp and the degree of work fatigue of the personnel. The task scheduling module is used to generate a quality inspection task allocation plan based on the task matching degree, and send the quality inspection task allocation plan to the corresponding quality inspection personnel for execution; The system is deployed on the server side, stores the status data of the quality inspection task agent and the QC agent in the database, and interfaces with the enterprise production management system to obtain the original task and personnel information.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multi-agent-based garment quality inspection task allocation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-agent-based garment quality inspection task allocation method as described in any one of claims 1 to 7.

Images